2015-01-27 01:45 AM
I'm trying to solve an issue that I'm facing with a high read latency on a NFS v4.1 exported volume.
The system is FAS8020 running cDOT 8.2.1.
The volume is situated in 14 disk (1 RG) SATA aggregate on node 2.
The volume is mounted by a Centos Linux machine supporting pNFS.
The volume is exported from vserver with lif running on interface group created from 2x 10Gbs interfaces.
The workload that we are evaluating is a time shifting IPTV. We are continously writing to the volume to record 30 TV channels.
When we try to read one TV channel we get high read latency.
I'm attaching some info about the configuration and statistics in a file.
Any ideas and recomendations will be appreciated.
Thanks in advance
2015-02-04 01:02 PM
First, I'd recommend moving to 8.2.3. That is the optimal release in 8.2.x for NFS.
Second, do you have a data LIF on every node in the cluster? This is a requirement for proper functionality of pNFS. Otherwise, you'd just be doing regular NFSv4.1 and possibly doing remote i/o.
TR-4063 and TR-4067 cover this in detail.
You can check to see if pNFS is indeed being used via:
::> set diag
::*> pnfs devices show
::*> pnfs devices mapping show
Third, is your identity management configured properly to service NFSv4.x ID domain requests?
Finally, what do you have set for nm-switch on the vserver?